Evolve reusable coding-agent skills from failed trajectories with EvoSkill

Mine failed agent runs for reusable skills, benchmark the candidates, and keep only the variants that improve a supported coding agent over your baseline.

Prerequisites

Python environment, a supported coding agent such as Claude Code, OpenCode, OpenHands, Goose, or Codex CLI, and a benchmark or task set to optimize against

Installation

Use the upstream install or setup path that matches your environment:

• uv sync
• pip install -e .
• brew install --cask claude-code # Claude Code
• brew install opencode # OpenCode (v1.4.0+)

Requirements and caveats from upstream:

• Python API
• Python 3.12+

Basic usage or getting-started notes:

• Also join us on Discord to discuss your experience, share suggestions, or show off your work!

• 🤖 Supported agents

• <table>

• Source: https://github.com/sentient-agi/EvoSkill

• Extracted from upstream docs: https://raw.githubusercontent.com/sentient-agi/EvoSkill/HEAD/README.md

Documentation

• https://github.com/sentient-agi/EvoSkill

Source

• Agent Skill Exchange